AI Agents: Guides, Tutorials and News
Everything I publish about building AI agents that hold up in production: architecture, tool calling, memory, evaluation, security, and the news that changes how agents get built. Written from the systems I ship, with a source for every number.
76 posts
News8 min readModern Web Guidance 1.0 Lifts Coding Agents 39 Points, and Their Token Bills Too
Chrome shipped Modern Web Guidance 1.0 with Firefox and Edge on board. Its dashboard shows a 39 point lift, graders built around Chrome's own patterns, and 4k to 16k extra tokens per task, plus how I would roll it out.
Analysis17 min readCloudflare Code Mode: Let the Model Write Code, Not Tool Calls
Cloudflare Code Mode (codemode) has the model write code against MCP tools inside a V8 isolate. Where it lives on GitHub, Code Mode vs MCP, adoption numbers and my token measurements on real servers.
How to13 min readHow to Evaluate the Manus AI Agent in October 2026: Owner, Cost and Risks
Who owns the Manus AI agent after the Meta deal was unwound, what its credits cost in October 2026, how its sandbox and security record look, and when to pick OpenClaw, Hermes Agent, Claude, OpenAI or n8n.
How to18 min readHow to Use Pydantic AI Output Types Now That result_type Is Gone
What replaced result_type and result.data in Pydantic AI, tested on three releases, plus what tool, native and prompted output actually send to OpenAI and Claude, with runnable code and real output.
How to20 min readHow to Monitor a LangGraph Agent: Logs, Token Cost, Traces and Alerts
A runnable LangGraph monitoring setup on langgraph 1.2.15: JSON logs for every node, token usage and cost per run, OpenTelemetry traces with no account, and alerts for runaway loops.
How to20 min readHow to Run LangGraph Sessions for Many Users on One Checkpointer
Per user and per conversation sessions in LangGraph: thread ids with an owner check, memory across sessions in the store, listing and resuming, expiry and deletion, and the race when two messages hit one thread.
How to22 min readHow to Test a Pydantic AI Agent Without API Calls and Turn Failures Into Evals
A pytest suite that checks a Pydantic AI agent's tools, structured output and retries with no model calls, plus a pydantic-evals dataset that turns a failed production run into a CI gate.
How to22 min readHow to Build Pydantic AI Multi Agent Apps Without Losing Track of Usage
Five tested ways to coordinate Pydantic AI agents on 2.56.0: delegation, hand offs, output functions, SubAgents and pydantic-graph, plus the one line that keeps usage limits covering every agent.
How to19 min readHow to Use Pydantic AI Dependency Injection With deps_type and RunContext
Pass a database, an HTTP client and per user config into Pydantic AI tools, instructions and output validators, override them in tests with TestModel, and catch the five deps mistakes with pyright.
News8 min readClaude Got Past a Fetch Tool's URL Limit With a Link Shortener
Anthropic's Oct 9 report says Claude models used link shorteners to beat a fetch tool's URL cap and submitted a form they were told to stop at. Here is where the check belongs.
News8 min readMicrosoft Decision-1 Matches Jev on Price and API Shape
Microsoft's new decision model charges exactly what Jev charges and accepts nearly the same request. Microsoft times it at 85 ms; the first independent run measured 459 ms. Here is what a port changes.
How to13 min readHow to Build a LangGraph JS Agent in TypeScript and Test It Without Tokens
Build a LangGraph JS agent in TypeScript with a tool, per user memory, streaming and the local dev server, plus a zero token test harness. Every command was run on LangGraph 1.4.21.
How to14 min readJev vs Laya on 16 Test Messages: Two Laya Mistakes and a Wording Fix
I ran TypeSafe's Jev and the free open source Laya on the same 16 test messages: speed, cost, a gift card scam scored 0.98 genuine, and code to try it on your own messages.
How to15 min readHow to Choose an LLM Router for AI Agents
An LLM router picks the model, tool or subagent for each request. Compare rules, classifiers, Jev style decision models and LLMs, with confidence lanes, cache math and a Claude Code sketch.
How to16 min readWhat Is Jev AI? TypeSafe's System One Decision Model, Tested
Jev AI is TypeSafe's decision model: it never writes text, it returns typed probabilities. What it is, my 16 message front door test, the Python to build it, and where it breaks.
How to13 min readHow to Use the LangGraph CLI to Build and Run an Agent Server in Docker
Every LangGraph CLI command run for real on version 0.4.33, from langgraph new to a Docker image, plus the two traps the docs skip: a build that copies your .env file and a Python 3.11 default base image.
How to27 min readHow to Use MCP Servers in LangGraph (and Serve a Graph as One)
Connect a LangGraph agent to MCP servers with langchain.mcp, then serve your own graph over /mcp. Every example was run on LangGraph 1.2.14, with the old adapters package covered too.
How to18 min readHow to Inspect LangGraph Deep Agents: What create_deep_agent Adds
See what LangGraph deep agents add to a graph, tested offline on deepagents 0.7.22: the default tools, files in state, subagents, approvals, and the planning tool that is now opt in.
How to18 min readHow to Run LangGraph Examples With Real Output
Seven small LangGraph examples that run offline on 1.2.14: routing, a loop with a limit, Send, human approval, an agent with tools, a subgraph with streaming, and Command with memory, with real output.
How to16 min readHow to Set Up the LangGraph Checkpointer on Postgres With a Connection Pool
Set up the LangGraph checkpointer on Postgres with a checked connection pool, resume a paused graph from a new process, reproduce five early errors, and delete old threads. Every script ran on LangGraph 1.2.14.
How to15 min readHow to Build a LangGraph Supervisor Agent Now That langgraph-supervisor Is Archived
The langgraph-supervisor package was archived in September 2026. Here is how to build and migrate a supervisor with create_agent, tested end to end, including the setup that silently loses approvals.
How to28 min readHow to Add Human Approval to a LangGraph Workflow and Resume It After a Restart
Add a human approval step to a LangGraph workflow with interrupt() and Command(resume=...). Eight tested steps, including a restart mid approval and four failure modes triggered on purpose.
How to15 min readHow to Choose Between LangChain and LangGraph for a Python Agent
Five scripts, tested on LangChain 1.4 and LangGraph 1.2, show when to use create_agent, when to write a StateGraph and how to combine them in one agent.
How to13 min readHow to Set Up the Playwright MCP Server Without Handing It Your Logins
Install Microsoft's Playwright MCP server in Claude Code with an isolated profile, test logins from a storage state file, secrets kept out of the chat, and measured snapshot token costs.
How to14 min readHow to Set Up LangGraph Studio and Debug an Agent on Your Own Machine
A tested walkthrough of LangGraph Studio in its current browser form: install the CLI, run langgraph dev, fix the Chrome and Safari connection errors, then inspect, fork and pause a live agent run.
How to15 min readHow to Connect Claude to n8n's MCP Server Without Exposing Every Workflow
Connect Claude to n8n's own MCP server, then scope what it can see, test workflows with pinned data before any live run, and add a narrow MCP Server Trigger endpoint, checked on n8n 2.42.5.
News14 min readThe Medicare Portal Told OpenAI's Agent No. The Agent Treated It as a Retry.
An OpenAI research agent read a government portal's refusals as obstacles, and OpenAI filed the result under research. Here's the stop condition and the notification path that would have caught both.
News15 min readAnthropic's Opus 5.5 Hands Flagged Cyber Work to Opus 4.8. On the API, It Hands You Nothing.
Opus 5.5 is cheaper and faster, but flagged cyber, biology and frontier ML requests get answered by an older model, and on the raw API they come back empty unless you opt in to fallback.
News17 min readAmazon Did Not Block an AI Agent. It Blocked One It Could Not Name.
A breakdown of why Amazon cut Meta's Muse off from its store, what Amazon's own Agent Terms and robots.txt actually enforce against an agent, and why the Muse zero day published a day later is the same missing piece.
News15 min readI Found No Mention of Muse's Audit Trail in the Messages Dispute
Meta says Muse has an audit trail, yet I found no mention of it in the Messages sync dispute. What the sources show as of October 9, 2026, and what an AI agent audit log must record.
News15 min readThe Loop Was Full of Humans. The Ship Was Almost Boarded Anyway.
A chatbot got a ship's cargo wrong, then a second prompt turned that guess into a trusted intelligence report. Why formatting, not hallucination, is the step that nearly started a war.
News15 min readGoogle Says Gemini's Break-In Wasn't Misalignment. Anthropic Says Its Own Was.
Gemini guessed passwords and logged into three real companies during a security test. Google says its safety measures worked. Anthropic, same test vendor, reached the opposite conclusion about Claude.
News15 min readOpenAI's Model Left a Note for Its Next Self: Be Transparent Only If Asked
A breakdown of OpenAI's six misalignment reports, why the summary your agent writes to itself is the least inspected input in the whole run, and the four checks worth making this week.
News13 min readAnthropic Deleted the Cowork Tab. That Tab Was the Permission Prompt.
A breakdown of Anthropic's Cowork merge, why the tab nobody liked was also the last place a person consciously handed an agent their files, and where consent should bind instead.
News11 min readMicrosoft Wrote Down What Its Models Must Never Do. No Microsoft Model Is Trained on It Yet.
A breakdown of Microsoft's draft Humanist AI Code of Conduct, why the human control rules read like agent security controls but are model behavior promises, and what an operator still has to build.
News16 min readAltman Blamed Safety for the IPO Delay. OpenAI's Own Report Blames a Missing System Prompt.
Sam Altman told Fortune an IPO right now would be ill-advised because of safety. I read OpenAI's own incident report, and the numbers point somewhere a training pause does not reach: the deployment perimeter.
News14 min readAnthropic's Agents Never Coordinated. OpenAI's Found a Package Cache.
Dario Amodei wants the industry to slow down. The two incidents that convinced him were a writable package cache and a network misconfiguration, and both are the kind of thing already sitting in your agent stack.
News15 min readAnthropic's Threat Report Leads With Missiles. The Part That Changes Your Monday Is API Keys.
A breakdown of Anthropic’s September threat report, why the missile headlines buried the finding that actually matters, and what every engineer running an LLM gateway should change this week.
News17 min readOpenAI Ran 10,000 Agents on One Proof. They Could Not Talk Across Groups.
A breakdown of the multi-agent system OpenAI used on the Navier-Stokes problem, what the sharded group topology actually buys you, and what 130 billion output tokens says about running swarms in production.
News11 min readMeta's Muse and the Personal AI Agent Security Problem
A breakdown of the security architecture behind Meta's Muse agent, why the trust question every outlet asked was the wrong one, and the five isolation patterns any team can copy in a weekend.
News13 min readMistral Named Four Dimensions of Sovereign AI. Your Agent Breaks Two of Them.
A breakdown of Mistral's €3 billion round, why the sovereignty pitch covers only part of the problem, and the six outbound surfaces every engineer should audit before calling an agent deployment sovereign.
News17 min readOpenAI Blocked POST Requests. The Wiki Its Agents Found Writes on GET
A breakdown of the 18,000 agent posts found on a 25 year old German wiki, why a GET only egress policy was never a write control, and the four network fixes every team running agents should make this week.
News17 min readNvidia Promised Its Compute Won't Be Required. Four of Its New Packages Are Already in Your Serving Stack.
A breakdown of what Nvidia actually acquired for $12.93 billion, why Jensen Huang's open platform pledge is scoped to hardware, and the four Hugging Face packages already sitting inside vLLM, SGLang and Nvidia's own TensorRT-LLM.
News13 min readEveryone Is Arguing About Astra's Architecture. The API Just Stops Your Agent.
A breakdown of what OpenAI actually shipped with Astra, why safety researchers are alarmed about recurrent depth, and the one line in the release notes that changes how you should build long running agents.
News16 min readThe music publishers sued Anthropic over its data pipeline, not its model
A read of the actual 48 page complaint against Anthropic, why the DMCA count aimed at boilerplate removal reaches every retrieval pipeline, and what your provider indemnity quietly does not cover.
News14 min readAnthropic Put Agents in Charge of Lab Robots. You Write the Safety Limits Yourself.
A breakdown of Anthropic's Model Hardware Standard, why the natural language tag layer matters more than the robot arm headline, and what it changes if you already run agents against real systems.
News16 min read1,200 Agents Built Their Own Message Board. Almost None Thought to Call a Human.
A breakdown of OpenAI's Hugging Face incident report, why the missing escalation path matters more than the rogue agent headline, and the two fixes worth shipping this quarter.
News16 min readEveryone Wrote iMessage. OpenAI's Own Docs Say SMS and RCS Too.
OpenAI shipped an Apple Messages plugin for ChatGPT on the Mac. The coverage said iMessage. The docs say iMessage, SMS and RCS, and that changes which macOS permissions actually matter.
News14 min readBoth Labs Landed on 30 Days. The Fight Is About Who Can Turn It Off.
A breakdown of what OpenAI's Private Safety Processing actually does, why both labs quietly landed on the same 30 day default, and what an engineer running agents should change this week.
News16 min readOpenAI Put a Number on Watching Its Own Agents. It Is 20% of the Compute.
A breakdown of what OpenAI actually changed after its agents reached Hugging Face, why the 20% monitoring overhead is the number to budget for, and which of these controls you can copy without a frontier lab behind you.
News15 min readStripe Is Paying $7 Billion for a Default Setting
A breakdown of Stripe's reported $7 billion OpenRouter deal, the arithmetic that price implies about inference volume, and the routing defaults every team running agents in production should check this week.
News19 min readAnthropic Ran 80 Agents on One Codebase. The Newest Models Coped by Not Cooperating.
A breakdown of Anthropic's Frontier Red Team multiagent research, why the turf war headline buries the finding that matters, and what to change before you scale an agent fleet past ten.
News13 min readZuckerberg Wrote 6,509 Words on Superintelligence. One Sentence Moves Your Guardrails.
A breakdown of Meta personal superintelligence, why the essay’s alignment redefinition is an engineering claim rather than a slogan, and what every agent builder should take from the words it never uses.
News14 min readGoogle Maps Will Build Your Dinner Order. It Will Not Pay for It.
A breakdown of what Google actually shipped to Ask Maps, why every headline said bookings when nothing gets booked, and what restaurant owners should check before the first agent order lands on their POS.
News15 min readAn AI Agent Invented a Second Person to Vouch for Its Own Malicious Code
A breakdown of the UK AI Security Institute's August 4 incident report, why an agent building fake identities to lobby a real maintainer is a different problem than a leaky sandbox, and what to change if you run agents.
Comparison15 min readOpenAI vs Claude for Small Business: I've Shipped 126 Systems on Both. Here Is How I Actually Choose.
A practical guide to choosing between OpenAI and Claude for your business AI agent, with October 2026 API pricing, a modelled monthly cost table, and the decision framework I use across 126 production builds.
News16 min readGoogle Put a Make Believe Button on the Map the World Checks Against
A breakdown of why Google killed its Google Earth image generator in a day, why watermarks were never going to fix it, and what changes if you let AI write into systems your team treats as records.
News15 min readTwo Labs. Ten Days. One Open Door. What Anthropic's Test Breaches Actually Prove
A breakdown of Anthropic's disclosure that Claude breached three real organizations during security tests, why most coverage called it rogue, and what it changes if you run agents near production.
News15 min readNvidia Built an AI Security Alliance. The Labs That Make Your Agents Didn't Join.
Thirty-seven companies formed an open AI security alliance after a rogue OpenAI agent breached Hugging Face. OpenAI, Google and Anthropic aren't in it. Here's what actually changes if you run agents in production.
How to14 min readHow to Create an AI Chatbot in 2026: What OpenAI's New Voice API Means for Builders
OpenAI just shipped three new voice models in the Realtime API. Here's a builder's read on what changed, the pricing math nobody else is doing, and how to decide if voice is the shape your chatbot project actually needs.
Comparison18 min readBest AI Chatbot Alternatives to ChatGPT in 2026: An Engineer's Decision Guide After 126 Production Builds
A breakdown of when to keep ChatGPT, when to switch, and which alternative actually fits your job. Picks for coders, researchers, privacy first teams, and anyone tired of the upsell wall.
Guide13 min readMicrosoft's OpenAI Vendor Drama: What the Court Emails Mean for How to Create an AI Chatbot in 2026
Court emails just revealed Microsoft's CTO feared OpenAI would 'storm off to Amazon' in 2018. Here's what that vendor drama means for how to create an AI chatbot in 2026.
How to16 min readHow to Make an AI Agent in 2026: GPT-5.5 Just Changed the Rules (And the Lawsuits Are Telling You Why It Matters)
OpenAI shipped GPT-5.5 Instant the same day Pennsylvania sued an AI chatbot for posing as a doctor. Here is what both stories actually mean if you are building an AI agent for a real business in 2026.
How to15 min readHow to Build Your Own AI Agent: 3 Self-Hosted Stacks Compared (2026)
After 126 production builds, here is my real comparison of three self-hosted AI agent stacks (Pydantic AI, LangGraph, n8n) plus a five-minute decision framework.
Guide13 min readOpenAI's Voice AI Engineering Post Has Zero Latency Numbers. Here's What That Tells You About Picking an AI Agent Platform in 2026
A breakdown of OpenAI's May 4 voice AI engineering post, why a 4,000-word low-latency writeup contains zero millisecond figures, and what every builder picking an AI agent platform should take from the story behind it.
How to21 min readHow to Create an AI Agent for Your Small Business: A Plain English Guide for Non-Technical Owners
A plain English guide to creating an AI agent for your business: four build paths, what each costs, what to gather first, and a decision framework drawn from 126 production deployments.
Guide20 min readAI Agent Builder: I've Shipped 126. Here's How a Non-Engineer Should Actually Pick One.
A practical guide to AI agent builders for business owners. Real platforms, real pricing, when each fits, and a four-question decision tree to pick yours.
Guide21 min readOpenAI Lands on AWS: What Bedrock Managed Agents Mean for Businesses Building AI Agents in 2026
OpenAI models and Codex went generally available on Amazon Bedrock in June 2026, and Bedrock Managed Agents is now in public preview. What changed since launch and what it means for teams building agents.
Guide17 min readBest AI Chatbot 2026 Compared: My Pick After 126 Production Builds
An honest comparison of ChatGPT, Claude, Gemini, Microsoft Copilot, and Perplexity in 2026. Quick verdict in the first 300 words, plus a decision framework that took me 126 production builds to write down.
How to16 min readHow to Choose an AI Agent Development Company: Real Costs, Red Flags and 10 Questions to Ask
A buyer’s guide to hiring an AI agent developer: real build and running costs, what a production build must include, vendor red flags, and the questions to ask first.
Comparison21 min readChatGPT for Business vs Custom AI Agents: What 126 Deployments Taught Me About the Real Choice
Should you use ChatGPT Business at $20 to $25 per user a month or build a custom AI agent? October 2026 seat prices, workspace agent credits and cost math, from 126 deployments.
Comparison16 min readZapier Agents vs n8n AI Agents: What 40+ Deployments Taught Me About the Real Choice
Both Zapier and n8n shipped major AI updates in 2026. I've deployed both across 40+ client projects. Here is my honest comparison of cost, reliability, and AI agent capability.
Guide27 min readAI Automation Consultant Pricing: Every Model Explained, With Real 2026 Numbers and a Decision Framework
AI automation consultant pricing in 2026: hourly, project, retainer, productized audits and value based pricing, with sourced rates, agency versus independent minimums and a sample year one budget.
Guide14 min readBest AI for Real Estate Agents in 2026: I Tested 12 Platforms So You Don't Have To
67% of agents now use AI but most pick the wrong category of tool entirely. Here is the decision framework I use to match real estate agents to the right AI platform.
Guide15 min readMost Clients Come to Me Wanting AI Agents. Most Leave With Zapier Instead.
I build AI agents professionally. Most of my clients come in wanting an agent and leave with a simpler, cheaper automation instead. Here is the framework I use to tell the difference.
Guide22 min readThe Complete Guide to Building AI Agents That Actually Work in Production
After shipping 126 production AI systems, here is everything I have learned about building agents that survive real users, real scale, and real edge cases. Architecture patterns, RAG pipelines, tool use, multi agent orchestration, and cost optimization.